Carnegie Mellon AI ‘Libratus’ Is Smoking Top Pro Poker Players With Tell Reads And Bluffing
These days, machine-learning along with artificial intelligence are becoming increasingly capable, and we're seeing their use applied to things we might not have immediately expected. Case in point: last month, we talked about how AI was being used to craft the most perfect beer.
But as wonderous as AI beer may sound, researchers at Carnegie Mellon University are going in a completely different direction, proving that a poker face is no match for advanced AI.
By design, poker is a simple game, but it's not at all simple on a competitive level. Even if you believe body language says your opponent isn't bluffing, they very well could still be, and there's a chance an AI opponent might be able to detect it more easily.
How an AI performs so well at these tasks is thanks to its learning capabilities, or deep-learning. CMU's "Libratus" AI program has become advanced to the point where it's just opened a lead over four of the top poker professional players in the world.
In Pittsburgh, a competition called "Brains vs Artificial Intelligence: Upping the Ante" is taking place. A staggering 120,000 hands of Heads-up, No-Limit Texas Hold'em poker are being played, a style that all four human players specialize in.
CMU's goal here isn't to ultimately beat these players a few times (though that may be a nice feather in their cap); it's to refine its algorithm. The best possible way to do that is to battle against real live players; not just some other form of AI. Ultimately, these players stand a chance to score a piece of a $200,000 prize, but for CMU, the research accomplished could create a new benchmark for AI.